(434f) Scheduling Multiproduct Pipeline Systems with RTN-Based Formulations

Pipelines provide a very efficient way of transporting large amounts of refined petroleum products (e.g. gasoline of different grades, diesel, LPG, jet fuel) for long distances. The pipeline system can comprise a single pipeline connecting an oil refinery to a depot, responsible for meeting the demands of a local consumer market, or be more complex, featuring a complex tree-like structure with a few refineries and depots.

The problem of optimally scheduling pipeline systems has received considerable attention in the last decade. While the earlier work of Rejowski and Pinto (2003) involved a pair of large-scale discrete-time MILP models for a pipeline with a single oil refinery and multiple depots, recent work has dealt with more accurate continuous-time approaches. Models by Cafaro and Cerdá (2004, 2009), Relvas et al. (2006) and MirHassani and Jahromi (2010) are conceptually similar in the sense that they explicitly deal with a set of batches and use variables to give their completion time, pumping length as well as volume.

In this work, we use a different approach, and treat the system as a continuous plant. The process is modeled as a Resource-Task Network to take advantage of previously developed general single time grid continuous-time formulations (Castro et al., 2004). In other words, the challenge is not with the mathematical model itself but with the process model. More specifically, a given pipeline segment can be viewed as a shared storage unit featuring a rather complex storage policy resulting from plug flow. We propose a pipeline segment module featuring filling, move and empty tasks, different pipeline states and volume resources. The latter relate to the actual product inside the pipeline as well as to the accumulated volume from the moment the product enters the pipeline, which can be due to itself entirely or to a mix of other products. Whenever the accumulated volume reaches the pipeline volume, a switch empty instantaneous event is triggered to start the discharge process. Switch fill & empty tasks are made sequence dependent to allow considering changeover costs and exclude forbidden product sequences. Note that a pumping batch may include the execution of multiple pumping tasks of the same product. The overall system will then comprise multiple pipeline segments as well as transfer from refinery and transfer to depots tasks.

The performance of the new approach is tested through the solution of a set of example problems taken from the literature. Preliminary results have shown that the higher generality of the new approach, which virtually enables the modeling of any structural arrangement systematically, is competitive when compared to the other approaches. Nevertheless, the number of events that can be handled effectively is roughly 7, making special tailored approaches a better option for simpler arrangements.